#pragma once // PublisherAdmin: the publisher model methods Discovery lists that the other // services in this module do not wrap: // - v1: serverStreamingPredict, the `v1/responses` family (create, compact, // get, delete) of a publisher; // - v1beta1: export, setPublisherModelConfig, enableModel, getIamPolicy, // the `v1/responses` family, and the v1beta1 form of every publisher // model method (generate, stream, tokens, predict, long running, raw // predict, invoke, embedContent, get). // Request bodies are checked against the call table (see EndpointAdmin). // Export, config changes and enableModel change the project: confirm=true. #include #include #include "google/gemini_types.h" #include "google/models/gemini.h" #include "google/models/vertex_service.h" #include "google/operation.h" namespace molo::google::models { class PublisherAdmin : public VertexService { public: using Sink = std::function; PublisherAdmin(std::shared_ptr client, VertexTarget target); // --- v1 -------------------------------------------------------------- void server_streaming_predict(const std::string& publisher, const std::string& model, const clients::body::Json& inputs, const clients::body::Json& parameters, const Sink& sink) const; // `v1/responses` of a publisher; version selects v1 or v1beta1. clients::body::Json responses_create(const std::string& publisher, const clients::body::Json& body, bool beta = false) const; clients::body::Json responses_compact(const std::string& publisher, const clients::body::Json& body, bool beta = false) const; clients::body::Json responses_get(const std::string& publisher, const std::string& response_id, bool beta = false) const; clients::body::Json responses_delete(const std::string& publisher, const std::string& response_id, bool confirm, bool beta = false) const; // --- v1beta1 --------------------------------------------------------- Operation export_model(const std::string& publisher, const std::string& model, const std::string& gcs_output_uri_prefix, bool confirm) const; Operation set_config(const std::string& publisher, const std::string& model, const clients::body::Json& publisher_model_config, const std::string& update_mask, bool confirm) const; clients::body::Json enable_model(const std::string& publisher, const std::string& model, const std::string& service, bool confirm) const; clients::body::Json get_iam_policy(const std::string& publisher, const std::string& model, int policy_version = 0) const; GenerateResponse beta_generate(const std::string& publisher, const std::string& model, const GenerateRequest& request) const; void beta_stream(const std::string& publisher, const std::string& model, const GenerateRequest& request, const std::function& on_chunk) const; CountTokensResult beta_count_tokens(const std::string& publisher, const std::string& model, const GenerateRequest& request) const; clients::body::Json beta_compute_tokens(const std::string& publisher, const std::string& model, const clients::body::Json& contents) const; clients::body::Json beta_get(const std::string& publisher, const std::string& model, bool full_view = false) const; // The remaining v1beta1 POST forms take their Discovery body as is: // call_id is one of beta.embed_content, beta.predict, beta.predict_long_running, // beta.fetch_predict_operation, beta.raw_predict, beta.server_streaming_predict. clients::body::Json beta_post(const std::string& call_id, const std::string& publisher, const std::string& model, const clients::body::Json& body) const; void beta_stream_raw_predict(const std::string& publisher, const std::string& model, const clients::body::Json& body, const Sink& sink) const; clients::body::Json beta_invoke(const std::string& publisher, const std::string& model, const std::string& invoke_id, const clients::body::Json& body) const; // Plans without sending (tests, call-table conformance). PlannedCall plan_call(const std::string& call_id, const CallPlanner::Args& args, const clients::body::Json* body = nullptr, const std::string& query = "") const; }; } // namespace molo::google::models